This project examines the cognitive mechanisms that are responsible for people's ability to understand argumentative prose. In prose of this kind, writers offer a series of sentences that lead to a final conclusion. In order to make sense of such a passage, readers must reconstruct the inferences that connect one sentence to the next, filling in implicit steps and placing explicit ones in proper relationship within the logical structure. The aim of this proposal is to design and test a theory of this comprehension process. The theory takes the form of a computational model that specifies how people interpret the sentences in an argumentative text, relate them logically to earlier sentences, and store them in memory. The model takes account of textual connectives (e.g., therefore, for, or however) and textual hints about inference methods (e.g., that the conclusion follows by a particular rule or from a previous result). Three groups of proposed experiments test the model's reasoning and memory assumptions by comparing its performance to that of adult subjects. The first group is based on the model's distinction between argumentative steps that are predictable and those that are unpredictable from previous steps; the experiments test this distinction by timing subjects as they comprehend and verify individual text lines. A second group of studies examines how the order of sentences in the text and the textual connectives in the sentences affect subjects' ability to identify the intended logical relations. The final set of experiments probes subjects' recall and recognition of alternative points of view within the global argument structure.